Fault diagnosis method for aero-engine with high dispersity cluster characteristics based on element engine

Abstract

The invention provides a fault diagnosis method for aero-engines with high-dispersity cluster characteristics based on element engines, and solves the problem that same gas path parameter values of engines of the same model are distributed dispersedly in a certain range due to manufacturing process level, assembly tolerance, working conditions, maintenance and the like of the aero-engines in actual engineering. Therefore, the health state of the engine is difficult to judge. Proposing and explaining high-dispersity cluster characteristics of aero-engine gas path parameters; proposing a meta-engine concept aiming at the characteristics, constructing a model of the meta-engine concept, and deriving an engine group with a high-dispersity cluster characteristic by utilizing the model; a wedge-leaf classifier and a novel classifier integration method are provided based on a meta-engine model, and fault diagnosis of an engine with high-dispersity cluster characteristics is achieved. According to the aero-engine fault diagnosis method provided by the invention, the generality and personality of engine gas path parameters are considered, the performance difference between the engine individuals is fully considered, and the aero-engine fault diagnosis method has universality for the cluster.

Description

Technical Field

[0001] The invention proposes a fault diagnosis method for an aero-engine with high dispersion cluster characteristics based on a meta-engine, and belongs to the field of intelligent aero-engine health management. Background Art

[0002] The prediction and health management technology of aircraft engines plays a very important role in evaluating and determining the working status and life status of the engine, as well as the safety and reliability of the entire aircraft. Identifying and eliminating faults or degradation problems caused by faults based on the acquired aircraft engine health status monitoring data is a core technology in health management.

[0003] When analyzing the field flight data of the same type of engine, it was found that the values ​​of the same gas path parameters between engines were dispersed within a certain range, and the differences could not be ignored, showing high-dispersion clustering. The possible reasons for the differences are: 1) tolerances in the production and assembly of engines; 2) different working hours, working conditions and execution tasks installed on different aircraft; 3) different noise pollution levels in the sensor data of each engine. The gas path parameter data of two engines may have different faults when they are similar, or they may have the same fault when they are not similar. Even the difference in parameter data of two engines with the same fault is greater than the difference in parameter data of the same engine under different faults, which cannot be identified by observing the engine data alone. Therefore, it is very meaningful to propose a new engine fault diagnosis method that can be applied to high-dispersion clustering characteristics.

[0004] The current mainstream aircraft engine fault diagnosis algorithms can be roughly divided into three categories: traditional fault prediction methods based on historical information, fault prediction methods based on physical models, and data-driven fault prediction methods. The data-driven method can avoid the establishment of complex mathematical models, which has attracted widespread attention. The patent with publication number CN112116101A discloses an aircraft engine fault diagnosis method based on a population-reduced core extreme learning machine, in which normal sub-components in the collected engine data within the full flight envelope are marked as positive labels, and vice versa. These data samples are used as training samples to train the population-reduced core extreme learning machine learning algorithm, and then the trained model is used to perform fault detection on various aircraft engine components. However, this method can only locate faults, but cannot identify the type of faults. In addition, the model obtained by correcting individuals can achieve good results on individuals, but cannot be directly applied to other individuals. The patent with the publication number CN118245924A discloses an aircraft engine fault diagnosis method based on domain distribution prediction adversarial migration. By constructing an adversarial migration model based on domain distribution prediction, the domain information of aircraft engine working data is characterized, and transfer learning of continuous domain and multi-dimensional domain is realized, which solves the problem that the aircraft engine lacks effective update data due to performance degradation and flight state changes during operation, resulting in a decrease in fault diagnosis accuracy. However, this method takes the working state of the engine as the focus of research, and verifies it within the full envelope range of one or several engines or throughout the life cycle, without verifying the universality of the proposed method on the same type of engine. Due to the high-dispersion cluster characteristics of the engine fleet, it is necessary to verify that the method has universality on the same type of engine. At present, there is no good method that can take into account the commonality and individual characteristics of the same type of engine and can be applied to the fault diagnosis model of the entire fleet. The proposed method of aircraft engine fault diagnosis based on the high-dispersion cluster characteristics of the meta-engine can effectively solve this problem. Summary of the invention

[0005] In view of the problems existing in the existing data-driven methods, the present invention proposes a fault diagnosis method for aircraft engines with high-dispersion cluster characteristics based on meta-engines, so as to realize fault diagnosis of engines with high-dispersion cluster characteristics.

[0006] To achieve the above objectives, the concept and technical solution of the present invention are implemented as follows:

[0007] The basic concept of the present invention is to propose and explain the high-dispersion cluster characteristics of aircraft engine gas path parameters, and explain the reasons why the gas path parameters of the same type of engine have both high dispersion and clustering; to propose the concept of meta-engine for the high-dispersion cluster characteristics of aircraft engine gas path parameters, and the meta-engine defines the generation specifications of important parameters of a certain type of engine; to build a meta-engine model based on the meta-engine, and use the model to derive an engine group with high-dispersion cluster characteristics; to propose a wedge leaf classifier and a new classifier integration method based on the meta-engine model, to partition the results of the basic classifier and analyze them in combination with the results of other parallel classifiers, so as to realize the fault diagnosis of engines with high-dispersion cluster characteristics. The model reflects the commonality and individuality of engine gas path parameters, fully considers the performance differences between individual engines, and is universal to the fleet.

[0008] Based on the above basic concept, the technical solution proposed by the present invention is a method for fault diagnosis of an aircraft engine based on high-dispersion cluster characteristics of a meta-engine, comprising the following steps:

[0009] Step 1: Propose and explain the high-dispersion cluster characteristics of aircraft engine gas path parameters;

[0010] Step 2: Propose a meta-engine concept based on the high-dispersion cluster characteristics of aircraft engine gas path parameters;

[0011] Step 3: Construct a meta-engine model, which consists of two parts: the initial engine and the data transformation module;

[0012] Step 4: Based on the meta-engine model, a wedge leaf classifier is proposed, which has a unified judgment standard;

[0013] Step 5: In order to enhance the classification ability of the wedge leaf classifier, a new classifier ensemble method is proposed to achieve fault diagnosis of engines with high dispersion cluster features.

[0014] Furthermore, in the process of explaining the high-dispersion cluster characteristics of the aircraft engine gas path parameters in step 1, 4 state parameters and 14 sensor parameters are included, specifically: the state parameters are altitude alt, flight Mach number Mach, throttle angle TRA and fan inlet total temperature T2; the sensor parameters are fuel flow Wf, physical fan speed Nf, physical core speed Nc, low-pressure compressor outlet total temperature T24, high-pressure compressor outlet total temperature T30, high-pressure turbine outlet total temperature T48, low-pressure turbine outlet total temperature T50, outer duct total pressure P15, fan inlet total pressure P2, fan outlet total pressure P21, low-pressure compressor outlet total pressure P24, high-pressure compressor outlet static pressure Ps30, combustion chamber outlet total pressure P40 and low-pressure turbine outlet total pressure P50.

[0015] The coefficient of dispersion, like the standard deviation and variance, can reflect the degree of data dispersion. It is defined as the ratio of the standard deviation to the mean:

[0016]

[0017] The coefficient of dispersion is a dimensionless quantity that can be used to compare two sets of data with different dimensions or data with different average values. The coefficient of dispersion is used to measure the dispersion of aircraft engines due to manufacturing processes, etc.

[0018] When an engine fails or degrades due to a failure, the engine will have a large difference from the rated working state when it leaves the factory. At this time, the parameters of the engine's gas path components measured by the sensor will change over time. The premise for gas path analysis and performance trend tracking is to obtain the engine's health status from these changing parameters. For the study of engine fault diagnosis, the relative data of parameter changes during engine operation is better used to evaluate the discreteness of engine performance. It is defined as:

[0019]

[0020] In the formula, Δx i It is the difference between the value of a parameter of the engine at the end of its service life and the value when it is just put into service. For all Δx i The average value of .

[0021] Furthermore, in step 2, the highly dispersed cluster characteristics of aircraft engine gas path parameters are inspired by the Western philosophy - metaphysics, and the concept of meta-engine is proposed. Metaphysics is an epistemological system based on the abstract, general and universal existence of human beings and the world. Meta can be understood as a self-description or behavior of things or concepts. The word meta is now used in many places, such as meta-learning, learning how to learn; metadata, data of data. Therefore, the meta-engine is the engine that produces the engine. The meta-engine defines the generation specifications of important parameters of a certain type of engine.

[0022] Furthermore, the meta-engine model in step 3 consists of two parts: the initial engine and the data transformation module. The initial engine is an individual engine, and the average value of the parameter data of the fleet is taken. The data transformation module consists of five data transformation methods. The specific steps are as follows:

[0023] Step 3.1: Add Gaussian noise, denoted as N(μ, σ 2 ), where μ, σ 2 are the expectation and variance of the Gaussian distribution respectively. Then the probability density function is:

[0024]

[0025] When the engine fleet is just shipped out of the factory, the tolerances caused by the production and assembly processes make the initial values ​​of the same parameters of different engines show a certain degree of dispersion, which can be regarded as a normal distribution. Adding Gaussian distribution can be regarded as adding different initial values ​​of parameters caused by tolerance to the initial engine.

[0026] Step 3.2: Data stretching, defined as:

[0027]

[0028] In an engine fleet, the same failure mode may show different degradation rates on different engines, and even when the performance degradation rate of each engine itself is not constant at different stages. Therefore, the gas path parameter data can be randomly stretched along the time axis to characterize different degradation rates.

[0029] Step 3.3: Data translation, defined as:

[0030]

[0031] Since the failure occurs at different times on different engines, the start time of performance degradation caused by the failure is also different. Random data transformation can be used to represent different degradation start times. That is, the data segment can be translated along the time axis to move certain degradation features contained in the data forward or backward.

[0032] Step 3.4: Data rotation, defined as:

[0033]

[0034] Where θ is the rotation angle.

[0035] Only some of the parameters of an aircraft engine can be measured, and the behavior of the system can only be observed from a limited perspective, so there are undiscovered relationships between different parameters of the engine. In order to improve the individual feature learning ability of the fleet, the multivariate time series is set as a window, and a small random rotation is performed around the center of the window to merge some features of adjacent parameters. Therefore, the random rotation of the data can be used to represent the relationship between different parameters.

[0036] Step 3.5: Data trimming. For two-dimensional data, it usually refers to parallel trimming of the four sides of the data. You can choose to randomly trim the bottom side to change the threshold of the parameter. The failure threshold of important engine parameters such as exhaust temperature has a certain degree of dispersion on different bodies. Random trimming of data can be used to characterize different thresholds. In addition, trimming the tail of time series data is also used to represent engine data at any stage of life.

[0037] Furthermore, in step 4, a wedge leaf classifier is proposed based on the meta-engine model. The wedge leaf classifier adopts the naive Bayes theory and regards all features as independent. All the acquired data of an engine are stretched or compressed into a fixed length along the time axis and defined as a window, which means that a window contains all the features of an engine. The branches and leaves of the Sphenopsida plants are all whorled leaf arrangements. A whole plant represents an engine cluster generated by a meta-engine. Each node of the stem represents an engine in the cluster. The whorled branches at the stem nodes and the whorled leaves on the branches are the characteristics of a single engine, which is vivid. Moreover, the three plants of wedge leaves, ferns and trees are from low to high levels, and the corresponding classifiers are also from simple to complex.

[0038] Furthermore, in order to enhance the classification ability of the wedge leaf classifier in step 5, a new classifier integration method is proposed. This method assigns different weights to each classifier in the cascade classifier based on the sensitivity of the classifier to recognize different features, thereby constructing multiple cascade classifiers. These cascade classifiers are trained in parallel. In addition to the wedge leaf classifier, the cascade classifier also includes a variance classifier and a trajectory matching classifier, which are defined as:

[0039] Variance classifier: E(p′ 2 )-E 2 (p′)>a*E(p 2 )-E 2 (p)

[0040] Where E is the expected value, p′ represents the derived engine window, p represents the engine window to be classified, and a is the coefficient.

[0041] Trajectory matching classifier:

[0042] Where S is the engine window to be classified; T is the derived engine window. The size of both windows is M×N. M represents the length of the time series, and N represents the type of parameter. mn and t mn Represent the elements in S and T respectively. and They represent the average values ​​of S and T respectively. The results are in the range of [-1, 1]. The closer R is to 1, the greater the correlation.

[0043] First, set the basic classifier and set the weight of the classifier according to the sensitivity of the classifier to the feature. The high sensitivity is set to a smaller value, and the low sensitivity is set to a larger value. The basic classifier is used to classify each engine to be classified, and then combined with the results of other parallel classifiers for analysis. The specific steps are as follows:

[0044] Step 5.1: Each type of fleet is derived from the initial engine into a corresponding fleet. Each fleet has the same number of engines;

[0045] Step 5.2: Calculate and compare the similarity between the test engine data and all engines of each fleet. If the similarity between the derived engine and the test engine reaches a certain standard, the derived engine is considered to have passed the classifier;

[0046] Step 5.3: Among all the engines that have passed the classification, if the number of engines belonging to one category is much greater than that of other categories, the test engine belongs to this category. If there is no significant difference, first select the category with the largest number of engines and record the number as the observed data of the test engine. For example, suppose the engine to be classified belongs to one of two types. Each engine derives one hundred engines, for a total of two hundred derived engines. In the end, eighty derived engines passed the classifier, of which seventy belonged to the first type and ten belonged to the second type. The engine to be tested is considered to belong to the first category. If forty-five belong to the first type and thirty-five belong to the second type, it is impossible to directly determine which category the engine to be tested belongs to. First record forty-five as a feature of the engine to be tested for the following analysis;

[0047] Step 5.4: Divide these engines with observed values ​​into several subsets, and then study each subset separately. The basis for division is the size of the observed values ​​of these engines. Take the observed value as the axis, take points in equal intervals on the axis, and take all the engines between the two points as a subset;

[0048] Step 5.5: Except for the basic classifier, the other cascade classifiers highlight different features by changing the weights of the classifiers. One of these cascade classifiers is selected according to the characteristics of each subset, and the selected cascade classifier is used to classify all samples in the subset. For each subset, after all samples pass through the selected parallel classifier, the observed values ​​obtained will change compared with the initial observed values. This is because the characteristics of different types of engines have different sensitivities to the classifier. The changes in the classifier settings cause different types of samples to produce different changes. For example, the observation data of one type becomes larger, while the observation data of the other type becomes smaller, clustered into two piles of data that can be linearly divided. Finally, the classification results of all intervals are combined to achieve the classification of all test engines.

[0049] Compared with the prior art, the present invention has the following beneficial effects:

[0050] In the present invention, in order to solve the problem that the flight data of the same type of engines are dispersed within a certain range, which brings difficulties to data analysis, a method for fault diagnosis of aircraft engines based on high-dispersion cluster characteristics of meta-engines is proposed. The high-dispersion cluster characteristics of aircraft engines are proposed and explained, and the influence of the dispersion of gas path parameters of the same type of engines on fault diagnosis is emphasized; the concept of meta-engine provides a way to obtain the common characteristics and individual characteristics of engines of the same type; the wedge leaf classifier uses the special shape of the wedge leaf to vividly express the time series data contained in the engine, and realizes rapid classification; the new classifier integration method is based on the divide-and-conquer idea, and uses the sensitivity of classifiers to identify different features to achieve accurate fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 It is a flow chart of a new method for fault diagnosis of a high-dispersion cluster characteristic engine provided by the present invention;

[0052] Figure 2 It is a schematic diagram of the meta-engine model provided by the present invention;

[0053] Figure 3 It is a schematic diagram of a data conversion module provided by the present invention;

[0054] Figure 4 It is a schematic diagram of a new classifier integration method provided by the present invention;

[0055] Figure 5 It is a schematic diagram of the output result and interval division of the basic classifier of the sample requiring further diagnosis provided by the present invention;

[0056] Figure 6 It is a schematic diagram of the d-interval diagnosis results using the new classifier integration method provided by the present invention. DETAILED DESCRIPTION

[0057] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0058] See also Figure 1 , which shows a flow chart of a method for fault diagnosis of an aircraft engine based on high-dispersion cluster characteristics of a meta-engine provided by the present invention, which specifically includes the following steps:

[0059] Step 1: The process of explaining the high-dispersion cluster characteristics of aircraft engine gas path parameters, including 4 state parameters and 14 sensor parameters, specifically: the state parameters are altitude alt, flight Mach number Mach, throttle angle TRA and fan inlet total temperature T2; sensor parameters are fuel flow Wf, physical fan speed Nf, physical core speed Nc, low-pressure compressor outlet total temperature T24, high-pressure compressor outlet total temperature T30, high-pressure turbine outlet total temperature T48, low-pressure turbine outlet total temperature T50, bypass duct total pressure P15, fan inlet total pressure P2, fan outlet total pressure P21, low-pressure compressor outlet total pressure P24, high-pressure compressor outlet static pressure Ps30, combustion chamber outlet total pressure P40 and low-pressure turbine outlet total pressure P50. In addition, there are two types of faults. The first type of fault is manifested as high-pressure compressor degradation, and the second type of fault is manifested as high-pressure compressor and fan degradation.

[0060] The coefficient of dispersion, like the standard deviation and variance, can reflect the degree of data dispersion. It is defined as the ratio of the standard deviation to the mean:

[0061]

[0062] The coefficient of dispersion is a dimensionless quantity that can be used to compare two sets of data with different dimensions or data with different average values. The coefficient of dispersion is used to measure the dispersion of aircraft engines due to manufacturing processes, etc. The coefficients of dispersion of 14 sensor parameters are shown in Table 1.

[0063] Table 1 Dispersion coefficient of sensor parameters

[0064]

[0065] When an engine fails or degrades due to a failure, the engine will have a large difference from the rated working state when it leaves the factory. At this time, the parameters of the engine's gas path components measured by the sensor will change over time. The premise for gas path analysis and performance trend tracking is to obtain the engine's health status from these changing parameters. For the study of engine fault diagnosis, the relative data of parameter changes during engine operation is better used to evaluate the discreteness of engine performance. It is defined as:

[0066]

[0067] In the formula, Δx i It is the difference between the value of a parameter of the engine at the end of its service life and the value when it is just put into service. For all Δx i The average value of .

[0068] Engine parameters c v,engine As shown in Table 2, it can be seen that the c of these parametersv,engine The maximum value is 1.33 and the minimum value is 0.26, and the degree of dispersion is relatively large.

[0069] Table 2 Sensor parameters c v,engine

[0070]

[0071] The main basis for gas path analysis and performance trends based on sensor parameters is the change of parameter data with the increase of service time, and the research object is also these changes. The dispersion of the initial value of the parameter and the dispersion of the subsequent time series data are large compared with the change Δx. For the parameter itself, the dispersion coefficient shows that the numerical distribution of a parameter is relatively concentrated, so the engine data presents a high dispersion cluster feature.

[0072] Step 2: The concept of meta-engine is proposed for the high-dispersion cluster characteristics of aircraft engine gas path parameters. This concept is inspired by metaphysics in Western philosophy. Metaphysics focuses on the abstract, general and universal epistemological system of people's existence in the world. Meta refers to the self-description or behavior of things or concepts. Nowadays, the term meta is widely used in various fields, such as meta-learning, which means learning how to learn; metadata means data about data. Therefore, meta-engine means the engine that produces the engine. Meta-engine defines the generation specifications of important parameters of a specific type of engine.

[0073] Step 3: Please refer to Figure 2 The meta-engine model consists of two parts: the initial engine and the data transformation module. The initial engine is an individual engine, and the average value of the group parameter data is taken. Figure 3 The data transformation module consists of five data transformation methods. The specific steps are as follows:

[0074] Step 3.1: Add Gaussian noise. The initial value dispersion of the fleet sensor data is about 0.01-1.8%, so let μ be 0 and σ 2 is 5;

[0075] Step 3.2: Data stretching: If the data is presented in the form of a time series, it is stretched on the time axis. The average life of the test engine is about 200 flight cycles, the shortest life is about 130 flight cycles, and the longest life is about 350 flight cycles. Therefore, the upper and lower limits of the stretching are set to half and twice the initial engine life.

[0076] Step 3.3: Data translation, if the data is presented in the form of time series, then it is translated on the time axis. The average life of the test engine is about 200 flight cycles, the shortest life is about 130 flight cycles, and the longest life is about 350 flight cycles. Therefore, the maximum period of left and right translation is set to about 100 flight cycles.

[0077] Step 3.4: Data rotation, by blending partial data from different sensors, learn the trends and relationships between parameters, that is, learn the lateral relationship between highly dispersed cluster engines. The rotation angle should not be too large, within 5 degrees.

[0078] Step 3.5: Crop the tail of the time series data of all parameters to simulate different thresholds of parameters or the data included from the service to different life stages of the engine. The cropping length can refer to the distribution of all engine parameter values ​​in the last cycle and the life range of the engine.

[0079] The above data conversion methods are randomly combined to form a data conversion module with a set probability of occurrence and a random occurrence order. Each initial engine randomly derives 100 engine data.

[0080] Step 4: Based on the meta-engine model, a wedge leaf classifier is proposed. The wedge leaf classifier uses the naive Bayesian theory and regards each feature as independent of each other. All the acquired data of an engine are stretched or compressed into a fixed length along the time axis and defined as a window, which means that a window contains all the features of an engine. The branches and leaves of the Sphenopsida plants are all whorled leaf arrangements. The whole plant represents an engine group generated by a meta-engine. Each node of the stem represents an engine in the group, and the whorled branches at the stem nodes and the whorled leaves on the branches are the characteristics of a single engine, which vividly presents this concept. In addition, the three plants of wedge leaves, ferns and trees are arranged from low to high levels, and the corresponding classifiers also develop from simple to complex.

[0081] Step 5: Please refer to Figure 4 , a new classifier ensemble method is used to realize the fault diagnosis of engines with high dispersion cluster features. Based on the sensitivity of classifiers to identify different features, the method assigns different weights to each classifier in the cascade classifier, thereby constructing multiple cascade classifiers. These cascade classifiers are trained in parallel. In addition to the wedge leaf classifier, the cascade classifier also contains a variance classifier and a trajectory matching classifier, which are defined as:

[0082] Variance classifier: E(p′ 2 )-E 2 (p′)>0.7*E(p 2 )-E 2 (p)

[0083] Where E is the expected value, p' represents the derived engine window, and p represents the engine window to be classified. If the variance of the derived engine window is greater than 70% of the variance of the engine window to be classified, the window passes the classifier.

[0084] Trajectory matching classifier:

[0085] Where S is the engine window to be classified; T is the derived engine window. The size of both windows is M×N. M represents the length of the time series, and N represents the type of parameter. mn and t mn Represent the elements in S and T respectively. and They represent the average values ​​of S and T respectively. The results are in the range of [-1, 1]. The closer R is to 1, the greater the correlation.

[0086] First, set the basic classifier and set the weight of the classifier according to the sensitivity of the classifier to the feature. The high sensitivity is set to a smaller value, and the low sensitivity is set to a larger value. The basic classifier is used to classify each engine to be classified, and then combined with the results of other parallel classifiers for analysis. The specific steps are as follows:

[0087] Step 5.1: Each type of fleet is derived from the initial engine into a corresponding fleet. Each fleet has the same number of engines;

[0088] Step 5.2: Calculate and compare the similarity between the test engine data and all engines of each fleet. If the similarity between the derived engine and the test engine reaches a certain standard, the derived engine is considered to have passed the classifier;

[0089] Step 5.3: In the classification results of the basic classifier, 48% of the engine samples can be strictly and clearly classified correctly, and the observations of the remaining engines are recorded. It is worth noting that these observations are almost all the number of the first type of derivative engines. That is to say, for each of the remaining engines to be tested, the number of derivative engines that pass the classifier is almost always slightly larger than the second type. This is explainable. Because the two engines involved are engines with different faults. The first fault is manifested as HPC degradation, and the second fault is manifested as HPC and fan degradation. Both faults include HPC degradation, and the second also has fan degradation. Fan degradation may give the engine some independent characteristics, or it may give the engine some characteristics caused by the combined effects of fan degradation and HPC degradation, and may even accelerate HPC degradation, making the characteristics of HPC degradation easier to capture, thereby causing the classifier to judge the second type of derivative engines as the first category. Therefore, for some test engines, the number of derivative engines that pass the classifier and belong to the first category may be greater;

[0090] Step 5.4: Divide the engines with observations into several subsets and study each subset separately. Figure 5 The basis for division is the size of the observed values ​​of these engines. Take the observed value as the axis, select points in equal intervals on the axis, and take all engines between the two points as a subset;

[0091] Step 5.5: In view of the fact that almost all the observations come from the first category and the specific analysis of these two types of engines, the setting of the parallel cascade classifier is mainly to adjust the parameters for the sensitive features related to the high-pressure compressor degradation. For example, it was found in the experiment that when the wedge blade classifier parameters are changed to allow more derivative engines of the two categories to enter the trajectory matching classifier, the trajectory matching classifier is more likely to classify the samples into the first category. Therefore, it is believed that the trajectory matching classifier is more sensitive to the degradation characteristics of the high-pressure compressor, while the wedge blade classifier has a certain ability to capture fan degradation. The parameters of the wedge blade classifier can be changed to observe the results of the trajectory matching classifier. Use the parallel cascade classifier to classify all samples in the subset again. Figure 6 The final classification results are shown using the d interval sample as an example, and the wrong classification is circled. Finally, the classification results of all intervals are combined to achieve fault diagnosis for all test engines. There are a total of 200 test engines, and the accuracy of fault diagnosis is 98%.

[0092] Regarding engine fault diagnosis, some existing methods include traditional classifiers, neural networks, and physical models. Compared with these traditional methods, the method provided by the present invention has the following advantages and characteristics. Traditional classifiers such as Boosting and Bagging are difficult to be used directly to solve the problem of this study, or will obtain results with low accuracy. Compared with traditional classifiers, this method takes into account the characteristics of high-dispersion cluster engines, and proposes a new method of integrating meta-engine models and classifiers to effectively solve the fault diagnosis problem of engine clusters. This method is more suitable for the object of this study and obtains accurate results. Neural networks have excellent performance in data processing and avoid the construction of complex mechanism models. However, neural networks are abstract and complex, and many phenomena cannot be well explained. This method combines engine knowledge and has good explanatory power for both the method and the results of verification. Research based on physical models can obtain accurate results based on engine mechanisms, but accurate physical models are difficult to construct. In addition, some physical models cannot be applied to multiple engines and need to be modified for each individual, which requires a lot of energy and time, while this method can avoid these problems. In general, this method has higher accuracy than traditional classifiers, better interpretability than neural networks, and is simpler and more generalizable than physical models.

[0093] The present invention is not limited to the above-mentioned embodiments. On the basis of the technical solution disclosed in the present invention, technicians in this field can make some simple modifications, equivalent changes and modifications to some of the technical features therein according to the disclosed technical content without creative labor, which all belong to the scope of the technical solution of the present invention.

Claims

1. A method for fault diagnosis of an aircraft engine based on high-dispersion cluster characteristics of a meta-engine, characterized in that: The specific steps include: Step 1: Propose and explain the high-dispersion cluster characteristics of aircraft engine gas path parameters; Step 2: Propose a meta-engine concept based on the high-dispersion cluster characteristics of aircraft engine gas path parameters; Step 3: Based on the concept of meta-engine, a meta-engine model is constructed, which consists of two parts: the initial engine and the data transformation module. The model is used to derive an engine group with high-dispersion cluster characteristics; Step 4: Based on the meta-engine model, a wedge leaf classifier is proposed, which has a unified judgment standard; Step 5: In order to enhance the classification ability of the wedge leaf classifier, a new classifier integration method is proposed. The results of the basic classifier are partitioned and then analyzed in combination with the results of other parallel classifiers to achieve fault diagnosis of engines with high dispersion cluster features.

2. The method for fault diagnosis of an aircraft engine based on a high-dispersion cluster feature of a meta-engine according to claim 1, characterized in that: The step 1 proposes and explains the high dispersion clustering characteristics of the air path parameters of the aircraft engine, and explains why the air path parameters of the same type of engine have both high dispersion and clustering. In statistics, a statistic that characterizes the degree of dispersion of a probability distribution is the coefficient of dispersion, which is defined as the standard deviation σ and the mean value. The ratio: The dispersion coefficient is used to measure the clustering of gas path parameters of engines of the same model and the level of engine manufacturing technology. Since the evaluation of the health of the engine is based on the change of the parameters of each gas path component over time, the discreteness of the engine parameters can be evaluated by using the relative data of the parameter changes during the engine operation, rather than the absolute values ​​of the parameters: In the formula, Δx i It is the difference between the value of a parameter of the engine at the end of its service life and the value when it is just put into service. For all Δx i The average value of .

3. The method for fault diagnosis of an aircraft engine based on a high-dispersion cluster feature of a meta-engine according to claim 1, characterized in that: The step 2 proposes the concept of a meta-engine for the high-dispersion cluster characteristics of aircraft engine gas path parameters. Meta can be understood as a self-description or behavior of a thing or concept. The meta-engine defines the generation specifications of important parameters of a certain type of engine.

4. The method for fault diagnosis of an aircraft engine based on a high-dispersion cluster feature of a meta-engine according to claim 1, characterized in that: The step 3 constructs a meta-engine model based on the meta-engine concept, which consists of two parts: the initial engine and the data transformation module. The initial engine is an individual engine, representing the commonality of the engine group; The data transformation module derives the individual characteristics of the cluster. The linear transformation, translation transformation and their combination of two-dimensional data can be expressed by a formula: In the formula, the left side of the equation is the transformed data, the first matrix on the right side of the equation is the transformation matrix, and the second one is the original data.

5. The method for fault diagnosis of an aircraft engine based on a high-dispersion cluster feature of a meta-engine according to claim 1, characterized in that: The step 4 proposes a wedge leaf classifier based on the meta-engine model. The classifier is inspired by the shape of the wedge leaf, combined with the engine data features, and adopts the naive Bayesian theory to achieve simple and fast preliminary classification.

6. The method for fault diagnosis of an aircraft engine based on a high-dispersion cluster feature of a meta-engine according to claim 1, characterized in that: In step 5, a new classifier integration method is proposed to enhance the classification ability of the wedge leaf classifier. The characteristics of engines in different states have different sensitivities to the classifier, and the different changes in the samples are observed by changing the settings of the classifier. The results of the basic classifier are partitioned, and then analyzed in combination with the results of other parallel classifiers to achieve fault diagnosis of engines with high dispersion cluster characteristics.

Images